2 * x86 FPU, MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI helpers
4 * Copyright (c) 2003 Fabrice Bellard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
22 #include "exec/helper-proto.h"
24 #include "qemu/host-utils.h"
25 #include "exec/cpu_ldst.h"
27 #define FPU_RC_MASK 0xc00
28 #define FPU_RC_NEAR 0x000
29 #define FPU_RC_DOWN 0x400
30 #define FPU_RC_UP 0x800
31 #define FPU_RC_CHOP 0xc00
33 #define MAXTAN 9223372036854775808.0
35 /* the following deal with x86 long double-precision numbers */
36 #define MAXEXPD 0x7fff
38 #define EXPD(fp) (fp.l.upper & 0x7fff)
39 #define SIGND(fp) ((fp.l.upper) & 0x8000)
40 #define MANTD(fp) (fp.l.lower)
41 #define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7fff)) | EXPBIAS
43 #define FPUS_IE (1 << 0)
44 #define FPUS_DE (1 << 1)
45 #define FPUS_ZE (1 << 2)
46 #define FPUS_OE (1 << 3)
47 #define FPUS_UE (1 << 4)
48 #define FPUS_PE (1 << 5)
49 #define FPUS_SF (1 << 6)
50 #define FPUS_SE (1 << 7)
51 #define FPUS_B (1 << 15)
55 #define floatx80_lg2 make_floatx80(0x3ffd, 0x9a209a84fbcff799LL)
56 #define floatx80_l2e make_floatx80(0x3fff, 0xb8aa3b295c17f0bcLL)
57 #define floatx80_l2t make_floatx80(0x4000, 0xd49a784bcd1b8afeLL)
59 static inline void fpush(CPUX86State
*env
)
61 env
->fpstt
= (env
->fpstt
- 1) & 7;
62 env
->fptags
[env
->fpstt
] = 0; /* validate stack entry */
65 static inline void fpop(CPUX86State
*env
)
67 env
->fptags
[env
->fpstt
] = 1; /* invalidate stack entry */
68 env
->fpstt
= (env
->fpstt
+ 1) & 7;
71 static inline floatx80
helper_fldt(CPUX86State
*env
, target_ulong ptr
)
75 temp
.l
.lower
= cpu_ldq_data(env
, ptr
);
76 temp
.l
.upper
= cpu_lduw_data(env
, ptr
+ 8);
80 static inline void helper_fstt(CPUX86State
*env
, floatx80 f
, target_ulong ptr
)
85 cpu_stq_data(env
, ptr
, temp
.l
.lower
);
86 cpu_stw_data(env
, ptr
+ 8, temp
.l
.upper
);
91 static inline double floatx80_to_double(CPUX86State
*env
, floatx80 a
)
98 u
.f64
= floatx80_to_float64(a
, &env
->fp_status
);
102 static inline floatx80
double_to_floatx80(CPUX86State
*env
, double a
)
110 return float64_to_floatx80(u
.f64
, &env
->fp_status
);
113 static void fpu_set_exception(CPUX86State
*env
, int mask
)
116 if (env
->fpus
& (~env
->fpuc
& FPUC_EM
)) {
117 env
->fpus
|= FPUS_SE
| FPUS_B
;
121 static inline floatx80
helper_fdiv(CPUX86State
*env
, floatx80 a
, floatx80 b
)
123 if (floatx80_is_zero(b
)) {
124 fpu_set_exception(env
, FPUS_ZE
);
126 return floatx80_div(a
, b
, &env
->fp_status
);
129 static void fpu_raise_exception(CPUX86State
*env
)
131 if (env
->cr
[0] & CR0_NE_MASK
) {
132 raise_exception(env
, EXCP10_COPR
);
134 #if !defined(CONFIG_USER_ONLY)
141 void helper_flds_FT0(CPUX86State
*env
, uint32_t val
)
149 FT0
= float32_to_floatx80(u
.f
, &env
->fp_status
);
152 void helper_fldl_FT0(CPUX86State
*env
, uint64_t val
)
160 FT0
= float64_to_floatx80(u
.f
, &env
->fp_status
);
163 void helper_fildl_FT0(CPUX86State
*env
, int32_t val
)
165 FT0
= int32_to_floatx80(val
, &env
->fp_status
);
168 void helper_flds_ST0(CPUX86State
*env
, uint32_t val
)
176 new_fpstt
= (env
->fpstt
- 1) & 7;
178 env
->fpregs
[new_fpstt
].d
= float32_to_floatx80(u
.f
, &env
->fp_status
);
179 env
->fpstt
= new_fpstt
;
180 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
183 void helper_fldl_ST0(CPUX86State
*env
, uint64_t val
)
191 new_fpstt
= (env
->fpstt
- 1) & 7;
193 env
->fpregs
[new_fpstt
].d
= float64_to_floatx80(u
.f
, &env
->fp_status
);
194 env
->fpstt
= new_fpstt
;
195 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
198 void helper_fildl_ST0(CPUX86State
*env
, int32_t val
)
202 new_fpstt
= (env
->fpstt
- 1) & 7;
203 env
->fpregs
[new_fpstt
].d
= int32_to_floatx80(val
, &env
->fp_status
);
204 env
->fpstt
= new_fpstt
;
205 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
208 void helper_fildll_ST0(CPUX86State
*env
, int64_t val
)
212 new_fpstt
= (env
->fpstt
- 1) & 7;
213 env
->fpregs
[new_fpstt
].d
= int64_to_floatx80(val
, &env
->fp_status
);
214 env
->fpstt
= new_fpstt
;
215 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
218 uint32_t helper_fsts_ST0(CPUX86State
*env
)
225 u
.f
= floatx80_to_float32(ST0
, &env
->fp_status
);
229 uint64_t helper_fstl_ST0(CPUX86State
*env
)
236 u
.f
= floatx80_to_float64(ST0
, &env
->fp_status
);
240 int32_t helper_fist_ST0(CPUX86State
*env
)
244 val
= floatx80_to_int32(ST0
, &env
->fp_status
);
245 if (val
!= (int16_t)val
) {
251 int32_t helper_fistl_ST0(CPUX86State
*env
)
255 val
= floatx80_to_int32(ST0
, &env
->fp_status
);
259 int64_t helper_fistll_ST0(CPUX86State
*env
)
263 val
= floatx80_to_int64(ST0
, &env
->fp_status
);
267 int32_t helper_fistt_ST0(CPUX86State
*env
)
271 val
= floatx80_to_int32_round_to_zero(ST0
, &env
->fp_status
);
272 if (val
!= (int16_t)val
) {
278 int32_t helper_fisttl_ST0(CPUX86State
*env
)
282 val
= floatx80_to_int32_round_to_zero(ST0
, &env
->fp_status
);
286 int64_t helper_fisttll_ST0(CPUX86State
*env
)
290 val
= floatx80_to_int64_round_to_zero(ST0
, &env
->fp_status
);
294 void helper_fldt_ST0(CPUX86State
*env
, target_ulong ptr
)
298 new_fpstt
= (env
->fpstt
- 1) & 7;
299 env
->fpregs
[new_fpstt
].d
= helper_fldt(env
, ptr
);
300 env
->fpstt
= new_fpstt
;
301 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
304 void helper_fstt_ST0(CPUX86State
*env
, target_ulong ptr
)
306 helper_fstt(env
, ST0
, ptr
);
309 void helper_fpush(CPUX86State
*env
)
314 void helper_fpop(CPUX86State
*env
)
319 void helper_fdecstp(CPUX86State
*env
)
321 env
->fpstt
= (env
->fpstt
- 1) & 7;
322 env
->fpus
&= ~0x4700;
325 void helper_fincstp(CPUX86State
*env
)
327 env
->fpstt
= (env
->fpstt
+ 1) & 7;
328 env
->fpus
&= ~0x4700;
333 void helper_ffree_STN(CPUX86State
*env
, int st_index
)
335 env
->fptags
[(env
->fpstt
+ st_index
) & 7] = 1;
338 void helper_fmov_ST0_FT0(CPUX86State
*env
)
343 void helper_fmov_FT0_STN(CPUX86State
*env
, int st_index
)
348 void helper_fmov_ST0_STN(CPUX86State
*env
, int st_index
)
353 void helper_fmov_STN_ST0(CPUX86State
*env
, int st_index
)
358 void helper_fxchg_ST0_STN(CPUX86State
*env
, int st_index
)
369 static const int fcom_ccval
[4] = {0x0100, 0x4000, 0x0000, 0x4500};
371 void helper_fcom_ST0_FT0(CPUX86State
*env
)
375 ret
= floatx80_compare(ST0
, FT0
, &env
->fp_status
);
376 env
->fpus
= (env
->fpus
& ~0x4500) | fcom_ccval
[ret
+ 1];
379 void helper_fucom_ST0_FT0(CPUX86State
*env
)
383 ret
= floatx80_compare_quiet(ST0
, FT0
, &env
->fp_status
);
384 env
->fpus
= (env
->fpus
& ~0x4500) | fcom_ccval
[ret
+ 1];
387 static const int fcomi_ccval
[4] = {CC_C
, CC_Z
, 0, CC_Z
| CC_P
| CC_C
};
389 void helper_fcomi_ST0_FT0(CPUX86State
*env
)
394 ret
= floatx80_compare(ST0
, FT0
, &env
->fp_status
);
395 eflags
= cpu_cc_compute_all(env
, CC_OP
);
396 eflags
= (eflags
& ~(CC_Z
| CC_P
| CC_C
)) | fcomi_ccval
[ret
+ 1];
400 void helper_fucomi_ST0_FT0(CPUX86State
*env
)
405 ret
= floatx80_compare_quiet(ST0
, FT0
, &env
->fp_status
);
406 eflags
= cpu_cc_compute_all(env
, CC_OP
);
407 eflags
= (eflags
& ~(CC_Z
| CC_P
| CC_C
)) | fcomi_ccval
[ret
+ 1];
411 void helper_fadd_ST0_FT0(CPUX86State
*env
)
413 ST0
= floatx80_add(ST0
, FT0
, &env
->fp_status
);
416 void helper_fmul_ST0_FT0(CPUX86State
*env
)
418 ST0
= floatx80_mul(ST0
, FT0
, &env
->fp_status
);
421 void helper_fsub_ST0_FT0(CPUX86State
*env
)
423 ST0
= floatx80_sub(ST0
, FT0
, &env
->fp_status
);
426 void helper_fsubr_ST0_FT0(CPUX86State
*env
)
428 ST0
= floatx80_sub(FT0
, ST0
, &env
->fp_status
);
431 void helper_fdiv_ST0_FT0(CPUX86State
*env
)
433 ST0
= helper_fdiv(env
, ST0
, FT0
);
436 void helper_fdivr_ST0_FT0(CPUX86State
*env
)
438 ST0
= helper_fdiv(env
, FT0
, ST0
);
441 /* fp operations between STN and ST0 */
443 void helper_fadd_STN_ST0(CPUX86State
*env
, int st_index
)
445 ST(st_index
) = floatx80_add(ST(st_index
), ST0
, &env
->fp_status
);
448 void helper_fmul_STN_ST0(CPUX86State
*env
, int st_index
)
450 ST(st_index
) = floatx80_mul(ST(st_index
), ST0
, &env
->fp_status
);
453 void helper_fsub_STN_ST0(CPUX86State
*env
, int st_index
)
455 ST(st_index
) = floatx80_sub(ST(st_index
), ST0
, &env
->fp_status
);
458 void helper_fsubr_STN_ST0(CPUX86State
*env
, int st_index
)
460 ST(st_index
) = floatx80_sub(ST0
, ST(st_index
), &env
->fp_status
);
463 void helper_fdiv_STN_ST0(CPUX86State
*env
, int st_index
)
468 *p
= helper_fdiv(env
, *p
, ST0
);
471 void helper_fdivr_STN_ST0(CPUX86State
*env
, int st_index
)
476 *p
= helper_fdiv(env
, ST0
, *p
);
479 /* misc FPU operations */
480 void helper_fchs_ST0(CPUX86State
*env
)
482 ST0
= floatx80_chs(ST0
);
485 void helper_fabs_ST0(CPUX86State
*env
)
487 ST0
= floatx80_abs(ST0
);
490 void helper_fld1_ST0(CPUX86State
*env
)
495 void helper_fldl2t_ST0(CPUX86State
*env
)
500 void helper_fldl2e_ST0(CPUX86State
*env
)
505 void helper_fldpi_ST0(CPUX86State
*env
)
510 void helper_fldlg2_ST0(CPUX86State
*env
)
515 void helper_fldln2_ST0(CPUX86State
*env
)
520 void helper_fldz_ST0(CPUX86State
*env
)
525 void helper_fldz_FT0(CPUX86State
*env
)
530 uint32_t helper_fnstsw(CPUX86State
*env
)
532 return (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
535 uint32_t helper_fnstcw(CPUX86State
*env
)
540 static void update_fp_status(CPUX86State
*env
)
544 /* set rounding mode */
545 switch (env
->fpuc
& FPU_RC_MASK
) {
548 rnd_type
= float_round_nearest_even
;
551 rnd_type
= float_round_down
;
554 rnd_type
= float_round_up
;
557 rnd_type
= float_round_to_zero
;
560 set_float_rounding_mode(rnd_type
, &env
->fp_status
);
561 switch ((env
->fpuc
>> 8) & 3) {
573 set_floatx80_rounding_precision(rnd_type
, &env
->fp_status
);
576 void helper_fldcw(CPUX86State
*env
, uint32_t val
)
579 update_fp_status(env
);
582 void helper_fclex(CPUX86State
*env
)
587 void helper_fwait(CPUX86State
*env
)
589 if (env
->fpus
& FPUS_SE
) {
590 fpu_raise_exception(env
);
594 void helper_fninit(CPUX86State
*env
)
611 void helper_fbld_ST0(CPUX86State
*env
, target_ulong ptr
)
619 for (i
= 8; i
>= 0; i
--) {
620 v
= cpu_ldub_data(env
, ptr
+ i
);
621 val
= (val
* 100) + ((v
>> 4) * 10) + (v
& 0xf);
623 tmp
= int64_to_floatx80(val
, &env
->fp_status
);
624 if (cpu_ldub_data(env
, ptr
+ 9) & 0x80) {
631 void helper_fbst_ST0(CPUX86State
*env
, target_ulong ptr
)
634 target_ulong mem_ref
, mem_end
;
637 val
= floatx80_to_int64(ST0
, &env
->fp_status
);
639 mem_end
= mem_ref
+ 9;
641 cpu_stb_data(env
, mem_end
, 0x80);
644 cpu_stb_data(env
, mem_end
, 0x00);
646 while (mem_ref
< mem_end
) {
652 v
= ((v
/ 10) << 4) | (v
% 10);
653 cpu_stb_data(env
, mem_ref
++, v
);
655 while (mem_ref
< mem_end
) {
656 cpu_stb_data(env
, mem_ref
++, 0);
660 void helper_f2xm1(CPUX86State
*env
)
662 double val
= floatx80_to_double(env
, ST0
);
664 val
= pow(2.0, val
) - 1.0;
665 ST0
= double_to_floatx80(env
, val
);
668 void helper_fyl2x(CPUX86State
*env
)
670 double fptemp
= floatx80_to_double(env
, ST0
);
673 fptemp
= log(fptemp
) / log(2.0); /* log2(ST) */
674 fptemp
*= floatx80_to_double(env
, ST1
);
675 ST1
= double_to_floatx80(env
, fptemp
);
678 env
->fpus
&= ~0x4700;
683 void helper_fptan(CPUX86State
*env
)
685 double fptemp
= floatx80_to_double(env
, ST0
);
687 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
690 fptemp
= tan(fptemp
);
691 ST0
= double_to_floatx80(env
, fptemp
);
694 env
->fpus
&= ~0x400; /* C2 <-- 0 */
695 /* the above code is for |arg| < 2**52 only */
699 void helper_fpatan(CPUX86State
*env
)
701 double fptemp
, fpsrcop
;
703 fpsrcop
= floatx80_to_double(env
, ST1
);
704 fptemp
= floatx80_to_double(env
, ST0
);
705 ST1
= double_to_floatx80(env
, atan2(fpsrcop
, fptemp
));
709 void helper_fxtract(CPUX86State
*env
)
715 if (floatx80_is_zero(ST0
)) {
716 /* Easy way to generate -inf and raising division by 0 exception */
717 ST0
= floatx80_div(floatx80_chs(floatx80_one
), floatx80_zero
,
724 expdif
= EXPD(temp
) - EXPBIAS
;
725 /* DP exponent bias */
726 ST0
= int32_to_floatx80(expdif
, &env
->fp_status
);
733 void helper_fprem1(CPUX86State
*env
)
735 double st0
, st1
, dblq
, fpsrcop
, fptemp
;
736 CPU_LDoubleU fpsrcop1
, fptemp1
;
738 signed long long int q
;
740 st0
= floatx80_to_double(env
, ST0
);
741 st1
= floatx80_to_double(env
, ST1
);
743 if (isinf(st0
) || isnan(st0
) || isnan(st1
) || (st1
== 0.0)) {
744 ST0
= double_to_floatx80(env
, 0.0 / 0.0); /* NaN */
745 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
753 expdif
= EXPD(fpsrcop1
) - EXPD(fptemp1
);
756 /* optimisation? taken from the AMD docs */
757 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
758 /* ST0 is unchanged */
763 dblq
= fpsrcop
/ fptemp
;
764 /* round dblq towards nearest integer */
766 st0
= fpsrcop
- fptemp
* dblq
;
768 /* convert dblq to q by truncating towards zero */
770 q
= (signed long long int)(-dblq
);
772 q
= (signed long long int)dblq
;
775 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
776 /* (C0,C3,C1) <-- (q2,q1,q0) */
777 env
->fpus
|= (q
& 0x4) << (8 - 2); /* (C0) <-- q2 */
778 env
->fpus
|= (q
& 0x2) << (14 - 1); /* (C3) <-- q1 */
779 env
->fpus
|= (q
& 0x1) << (9 - 0); /* (C1) <-- q0 */
781 env
->fpus
|= 0x400; /* C2 <-- 1 */
782 fptemp
= pow(2.0, expdif
- 50);
783 fpsrcop
= (st0
/ st1
) / fptemp
;
784 /* fpsrcop = integer obtained by chopping */
785 fpsrcop
= (fpsrcop
< 0.0) ?
786 -(floor(fabs(fpsrcop
))) : floor(fpsrcop
);
787 st0
-= (st1
* fpsrcop
* fptemp
);
789 ST0
= double_to_floatx80(env
, st0
);
792 void helper_fprem(CPUX86State
*env
)
794 double st0
, st1
, dblq
, fpsrcop
, fptemp
;
795 CPU_LDoubleU fpsrcop1
, fptemp1
;
797 signed long long int q
;
799 st0
= floatx80_to_double(env
, ST0
);
800 st1
= floatx80_to_double(env
, ST1
);
802 if (isinf(st0
) || isnan(st0
) || isnan(st1
) || (st1
== 0.0)) {
803 ST0
= double_to_floatx80(env
, 0.0 / 0.0); /* NaN */
804 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
812 expdif
= EXPD(fpsrcop1
) - EXPD(fptemp1
);
815 /* optimisation? taken from the AMD docs */
816 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
817 /* ST0 is unchanged */
822 dblq
= fpsrcop
/ fptemp
; /* ST0 / ST1 */
823 /* round dblq towards zero */
824 dblq
= (dblq
< 0.0) ? ceil(dblq
) : floor(dblq
);
825 st0
= fpsrcop
- fptemp
* dblq
; /* fpsrcop is ST0 */
827 /* convert dblq to q by truncating towards zero */
829 q
= (signed long long int)(-dblq
);
831 q
= (signed long long int)dblq
;
834 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
835 /* (C0,C3,C1) <-- (q2,q1,q0) */
836 env
->fpus
|= (q
& 0x4) << (8 - 2); /* (C0) <-- q2 */
837 env
->fpus
|= (q
& 0x2) << (14 - 1); /* (C3) <-- q1 */
838 env
->fpus
|= (q
& 0x1) << (9 - 0); /* (C1) <-- q0 */
840 int N
= 32 + (expdif
% 32); /* as per AMD docs */
842 env
->fpus
|= 0x400; /* C2 <-- 1 */
843 fptemp
= pow(2.0, (double)(expdif
- N
));
844 fpsrcop
= (st0
/ st1
) / fptemp
;
845 /* fpsrcop = integer obtained by chopping */
846 fpsrcop
= (fpsrcop
< 0.0) ?
847 -(floor(fabs(fpsrcop
))) : floor(fpsrcop
);
848 st0
-= (st1
* fpsrcop
* fptemp
);
850 ST0
= double_to_floatx80(env
, st0
);
853 void helper_fyl2xp1(CPUX86State
*env
)
855 double fptemp
= floatx80_to_double(env
, ST0
);
857 if ((fptemp
+ 1.0) > 0.0) {
858 fptemp
= log(fptemp
+ 1.0) / log(2.0); /* log2(ST + 1.0) */
859 fptemp
*= floatx80_to_double(env
, ST1
);
860 ST1
= double_to_floatx80(env
, fptemp
);
863 env
->fpus
&= ~0x4700;
868 void helper_fsqrt(CPUX86State
*env
)
870 if (floatx80_is_neg(ST0
)) {
871 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
874 ST0
= floatx80_sqrt(ST0
, &env
->fp_status
);
877 void helper_fsincos(CPUX86State
*env
)
879 double fptemp
= floatx80_to_double(env
, ST0
);
881 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
884 ST0
= double_to_floatx80(env
, sin(fptemp
));
886 ST0
= double_to_floatx80(env
, cos(fptemp
));
887 env
->fpus
&= ~0x400; /* C2 <-- 0 */
888 /* the above code is for |arg| < 2**63 only */
892 void helper_frndint(CPUX86State
*env
)
894 ST0
= floatx80_round_to_int(ST0
, &env
->fp_status
);
897 void helper_fscale(CPUX86State
*env
)
899 if (floatx80_is_any_nan(ST1
)) {
902 int n
= floatx80_to_int32_round_to_zero(ST1
, &env
->fp_status
);
903 ST0
= floatx80_scalbn(ST0
, n
, &env
->fp_status
);
907 void helper_fsin(CPUX86State
*env
)
909 double fptemp
= floatx80_to_double(env
, ST0
);
911 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
914 ST0
= double_to_floatx80(env
, sin(fptemp
));
915 env
->fpus
&= ~0x400; /* C2 <-- 0 */
916 /* the above code is for |arg| < 2**53 only */
920 void helper_fcos(CPUX86State
*env
)
922 double fptemp
= floatx80_to_double(env
, ST0
);
924 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
927 ST0
= double_to_floatx80(env
, cos(fptemp
));
928 env
->fpus
&= ~0x400; /* C2 <-- 0 */
929 /* the above code is for |arg| < 2**63 only */
933 void helper_fxam_ST0(CPUX86State
*env
)
940 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
942 env
->fpus
|= 0x200; /* C1 <-- 1 */
945 /* XXX: test fptags too */
947 if (expdif
== MAXEXPD
) {
948 if (MANTD(temp
) == 0x8000000000000000ULL
) {
949 env
->fpus
|= 0x500; /* Infinity */
951 env
->fpus
|= 0x100; /* NaN */
953 } else if (expdif
== 0) {
954 if (MANTD(temp
) == 0) {
955 env
->fpus
|= 0x4000; /* Zero */
957 env
->fpus
|= 0x4400; /* Denormal */
964 void helper_fstenv(CPUX86State
*env
, target_ulong ptr
, int data32
)
966 int fpus
, fptag
, exp
, i
;
970 fpus
= (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
972 for (i
= 7; i
>= 0; i
--) {
974 if (env
->fptags
[i
]) {
977 tmp
.d
= env
->fpregs
[i
].d
;
980 if (exp
== 0 && mant
== 0) {
983 } else if (exp
== 0 || exp
== MAXEXPD
984 || (mant
& (1LL << 63)) == 0) {
985 /* NaNs, infinity, denormal */
992 cpu_stl_data(env
, ptr
, env
->fpuc
);
993 cpu_stl_data(env
, ptr
+ 4, fpus
);
994 cpu_stl_data(env
, ptr
+ 8, fptag
);
995 cpu_stl_data(env
, ptr
+ 12, 0); /* fpip */
996 cpu_stl_data(env
, ptr
+ 16, 0); /* fpcs */
997 cpu_stl_data(env
, ptr
+ 20, 0); /* fpoo */
998 cpu_stl_data(env
, ptr
+ 24, 0); /* fpos */
1001 cpu_stw_data(env
, ptr
, env
->fpuc
);
1002 cpu_stw_data(env
, ptr
+ 2, fpus
);
1003 cpu_stw_data(env
, ptr
+ 4, fptag
);
1004 cpu_stw_data(env
, ptr
+ 6, 0);
1005 cpu_stw_data(env
, ptr
+ 8, 0);
1006 cpu_stw_data(env
, ptr
+ 10, 0);
1007 cpu_stw_data(env
, ptr
+ 12, 0);
1011 void helper_fldenv(CPUX86State
*env
, target_ulong ptr
, int data32
)
1016 env
->fpuc
= cpu_lduw_data(env
, ptr
);
1017 fpus
= cpu_lduw_data(env
, ptr
+ 4);
1018 fptag
= cpu_lduw_data(env
, ptr
+ 8);
1020 env
->fpuc
= cpu_lduw_data(env
, ptr
);
1021 fpus
= cpu_lduw_data(env
, ptr
+ 2);
1022 fptag
= cpu_lduw_data(env
, ptr
+ 4);
1024 env
->fpstt
= (fpus
>> 11) & 7;
1025 env
->fpus
= fpus
& ~0x3800;
1026 for (i
= 0; i
< 8; i
++) {
1027 env
->fptags
[i
] = ((fptag
& 3) == 3);
1032 void helper_fsave(CPUX86State
*env
, target_ulong ptr
, int data32
)
1037 helper_fstenv(env
, ptr
, data32
);
1039 ptr
+= (14 << data32
);
1040 for (i
= 0; i
< 8; i
++) {
1042 helper_fstt(env
, tmp
, ptr
);
1060 void helper_frstor(CPUX86State
*env
, target_ulong ptr
, int data32
)
1065 helper_fldenv(env
, ptr
, data32
);
1066 ptr
+= (14 << data32
);
1068 for (i
= 0; i
< 8; i
++) {
1069 tmp
= helper_fldt(env
, ptr
);
1075 #if defined(CONFIG_USER_ONLY)
1076 void cpu_x86_fsave(CPUX86State
*env
, target_ulong ptr
, int data32
)
1078 helper_fsave(env
, ptr
, data32
);
1081 void cpu_x86_frstor(CPUX86State
*env
, target_ulong ptr
, int data32
)
1083 helper_frstor(env
, ptr
, data32
);
1087 void helper_fxsave(CPUX86State
*env
, target_ulong ptr
, int data64
)
1089 int fpus
, fptag
, i
, nb_xmm_regs
;
1093 /* The operand must be 16 byte aligned */
1095 raise_exception(env
, EXCP0D_GPF
);
1098 fpus
= (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
1100 for (i
= 0; i
< 8; i
++) {
1101 fptag
|= (env
->fptags
[i
] << i
);
1103 cpu_stw_data(env
, ptr
, env
->fpuc
);
1104 cpu_stw_data(env
, ptr
+ 2, fpus
);
1105 cpu_stw_data(env
, ptr
+ 4, fptag
^ 0xff);
1106 #ifdef TARGET_X86_64
1108 cpu_stq_data(env
, ptr
+ 0x08, 0); /* rip */
1109 cpu_stq_data(env
, ptr
+ 0x10, 0); /* rdp */
1113 cpu_stl_data(env
, ptr
+ 0x08, 0); /* eip */
1114 cpu_stl_data(env
, ptr
+ 0x0c, 0); /* sel */
1115 cpu_stl_data(env
, ptr
+ 0x10, 0); /* dp */
1116 cpu_stl_data(env
, ptr
+ 0x14, 0); /* sel */
1120 for (i
= 0; i
< 8; i
++) {
1122 helper_fstt(env
, tmp
, addr
);
1126 if (env
->cr
[4] & CR4_OSFXSR_MASK
) {
1127 /* XXX: finish it */
1128 cpu_stl_data(env
, ptr
+ 0x18, env
->mxcsr
); /* mxcsr */
1129 cpu_stl_data(env
, ptr
+ 0x1c, 0x0000ffff); /* mxcsr_mask */
1130 if (env
->hflags
& HF_CS64_MASK
) {
1136 /* Fast FXSAVE leaves out the XMM registers */
1137 if (!(env
->efer
& MSR_EFER_FFXSR
)
1138 || (env
->hflags
& HF_CPL_MASK
)
1139 || !(env
->hflags
& HF_LMA_MASK
)) {
1140 for (i
= 0; i
< nb_xmm_regs
; i
++) {
1141 cpu_stq_data(env
, addr
, env
->xmm_regs
[i
].XMM_Q(0));
1142 cpu_stq_data(env
, addr
+ 8, env
->xmm_regs
[i
].XMM_Q(1));
1149 void helper_fxrstor(CPUX86State
*env
, target_ulong ptr
, int data64
)
1151 int i
, fpus
, fptag
, nb_xmm_regs
;
1155 /* The operand must be 16 byte aligned */
1157 raise_exception(env
, EXCP0D_GPF
);
1160 env
->fpuc
= cpu_lduw_data(env
, ptr
);
1161 fpus
= cpu_lduw_data(env
, ptr
+ 2);
1162 fptag
= cpu_lduw_data(env
, ptr
+ 4);
1163 env
->fpstt
= (fpus
>> 11) & 7;
1164 env
->fpus
= fpus
& ~0x3800;
1166 for (i
= 0; i
< 8; i
++) {
1167 env
->fptags
[i
] = ((fptag
>> i
) & 1);
1171 for (i
= 0; i
< 8; i
++) {
1172 tmp
= helper_fldt(env
, addr
);
1177 if (env
->cr
[4] & CR4_OSFXSR_MASK
) {
1178 /* XXX: finish it */
1179 cpu_set_mxcsr(env
, cpu_ldl_data(env
, ptr
+ 0x18));
1180 /* cpu_ldl_data(env, ptr + 0x1c); */
1181 if (env
->hflags
& HF_CS64_MASK
) {
1187 /* Fast FXRESTORE leaves out the XMM registers */
1188 if (!(env
->efer
& MSR_EFER_FFXSR
)
1189 || (env
->hflags
& HF_CPL_MASK
)
1190 || !(env
->hflags
& HF_LMA_MASK
)) {
1191 for (i
= 0; i
< nb_xmm_regs
; i
++) {
1192 env
->xmm_regs
[i
].XMM_Q(0) = cpu_ldq_data(env
, addr
);
1193 env
->xmm_regs
[i
].XMM_Q(1) = cpu_ldq_data(env
, addr
+ 8);
1200 void cpu_get_fp80(uint64_t *pmant
, uint16_t *pexp
, floatx80 f
)
1205 *pmant
= temp
.l
.lower
;
1206 *pexp
= temp
.l
.upper
;
1209 floatx80
cpu_set_fp80(uint64_t mant
, uint16_t upper
)
1213 temp
.l
.upper
= upper
;
1214 temp
.l
.lower
= mant
;
1219 /* XXX: optimize by storing fptt and fptags in the static cpu state */
1221 #define SSE_DAZ 0x0040
1222 #define SSE_RC_MASK 0x6000
1223 #define SSE_RC_NEAR 0x0000
1224 #define SSE_RC_DOWN 0x2000
1225 #define SSE_RC_UP 0x4000
1226 #define SSE_RC_CHOP 0x6000
1227 #define SSE_FZ 0x8000
1229 void cpu_set_mxcsr(CPUX86State
*env
, uint32_t mxcsr
)
1235 /* set rounding mode */
1236 switch (mxcsr
& SSE_RC_MASK
) {
1239 rnd_type
= float_round_nearest_even
;
1242 rnd_type
= float_round_down
;
1245 rnd_type
= float_round_up
;
1248 rnd_type
= float_round_to_zero
;
1251 set_float_rounding_mode(rnd_type
, &env
->sse_status
);
1253 /* set denormals are zero */
1254 set_flush_inputs_to_zero((mxcsr
& SSE_DAZ
) ? 1 : 0, &env
->sse_status
);
1256 /* set flush to zero */
1257 set_flush_to_zero((mxcsr
& SSE_FZ
) ? 1 : 0, &env
->fp_status
);
1260 void helper_ldmxcsr(CPUX86State
*env
, uint32_t val
)
1262 cpu_set_mxcsr(env
, val
);
1265 void helper_enter_mmx(CPUX86State
*env
)
1268 *(uint32_t *)(env
->fptags
) = 0;
1269 *(uint32_t *)(env
->fptags
+ 4) = 0;
1272 void helper_emms(CPUX86State
*env
)
1274 /* set to empty state */
1275 *(uint32_t *)(env
->fptags
) = 0x01010101;
1276 *(uint32_t *)(env
->fptags
+ 4) = 0x01010101;
1280 void helper_movq(CPUX86State
*env
, void *d
, void *s
)
1282 *(uint64_t *)d
= *(uint64_t *)s
;
1286 #include "ops_sse.h"
1289 #include "ops_sse.h"